I've just reclaimed my old Q6600 rig back that the kids no longer use and stuck 2 of my old 9800GT's back in it, plus modded the case that it's now while I was at it.

I installed BOINC 7.2.33 (no I will not be updating my main rigs to this version) and installed Lunatics.

This rig.

Now under the OpenCL entries for Astropulse in my app_info.xml if I change the entries
<avg_ncpus>0.04</avg_ncpus> to 1
and
<max_ncpus>0.2</max_ncpus> to 1
will this automatically reserve CPU cores for those Astropulse tasks running on the GPUs so that I don't have to reserve cores in my Local Preferences and when I only have MultiBeam work all 4 cores crunch without resorting to making an app_config.xml?

Cheers.

Every AP running on your graphics cards will reserve a CPU core in that case. Just be aware that if you are running 2 WUs per graphics card (is that possible with those?), in the rare case you are running 4 APs at once on them, no MBs will run on the CPU.

In my case (4 cores, 2 cards running 3 WUs each), I also must have an app_config.xml to prevent more than 4 APs running simultaneously.

---

I've just reclaimed my old Q6600 rig back that the kids no longer use and stuck 2 of my old 9800GT's back in it, plus modded the case that it's now while I was at it.

I installed BOINC 7.2.33 (no I will not be updating my main rigs to this version) and installed Lunatics.

This rig.

Now under the OpenCL entries for Astropulse in my app_info.xml if I change the entries
<avg_ncpus>0.04</avg_ncpus> to 1
and
<max_ncpus>0.2</max_ncpus> to 1
will this automatically reserve CPU cores for those Astropulse tasks running on the GPUs so that I don't have to reserve cores in my Local Preferences and when I only have MultiBeam work all 4 cores crunch without resorting to making an app_config.xml?

Cheers.

Yes, that should work as long as the multiprocessor setting is 100%.
But, why are you running APs on those G92 cards? Those cards run APs poorly. Also, you are using a driver that will consume almost an entire CPU core whereas driver 266.58 will use much less CPU when running APs. My G92s produce a higher RAC when running CUDAs. I also found that CUDA 3.2 works slightly better in XP whereas CUDA 2.3 works better in Windows 8.1. If I had that rig, I'd run CUDA 2.3s on the cards and APs on the OCed CPU...

Every AP running on your graphics cards will reserve a CPU core in that case. Just be aware that if you are running 2 WUs per graphics card (is that possible with those?), in the rare case you are running 4 APs at once on them, no MBs will run on the CPU.

In my case (4 cores, 2 cards running 3 WUs each), I also must have an app_config.xml to prevent more than 4 APs running simultaneously.

Thanks for the reassurance with that as that's what I thought, but I wanted to be sure 1st.

No, just a single AP workunit on each is all these old cards will do (you can run 2 MultiBeams on them, but it's not worth the effort).

Cheers.

Yes, that should work as long as the multiprocessor setting is 100%.
But, why are you running APs on those G92 cards? Those cards run APs poorly. Also, you are using a driver that will consume almost an entire CPU core whereas driver 266.58 will use much less CPU when running APs. My G92s produce a higher RAC when running CUDAs. I also found that CUDA 3.2 works slightly better in XP whereas CUDA 2.3 works better in Windows 8.1. If I had that rig, I'd run CUDA 2.3s on the cards and APs on the OCed CPU...

It's just an experiment in progress (I can't do much damage with them). ;-)

I'll let them run for a week with that driver (as it was already on the system with a 9800GTX+) and then I'll reload the 266.58 driver on it (I still have that here).

This rig also use to be my main rig with those same 9800GT's in it (low power jobs that draw all their power from the PCI-e slots, just down clocked from standard versions, but not the green power jobs that use the 9800mGT cores).

It also has been Win7 32-bit for a few years now and CUDA 3.2 was the best for it before I retired it to the kids about 5mths back with that 266.58 driver (it did get updated with the 2 GTX 660's that now reside in my 3570K rig just before then which is why it's now running the driver that it currently is).

Cheers.

It appears we'll have a nice little test going on. I plan on moving my newly acquired Q9400 to the Win 8 host with the two low end ATI cards and moving the two G92 cards to the XP host with the Dual core version of the Q6600. So, it will be close to the same machine, less two CPU cores. It might be a few days before the change though.

It appears we'll have a nice little test going on. I plan on moving my newly acquired Q9400 to the Win 8 host with the two low end ATI cards and moving the two G92 cards to the XP host with the Dual core version of the Q6600. So, it will be close to the same machine, less two CPU cores. It might be a few days before the change though.

My Q6600 is a SLACR running at a mild overclock of 3GHz, but once I upgrade the GPUs 1 of my main rigs I'll likely have the 2 GTX 550Ti's from the 2500K in it replacing the 9800GT's.

Well the 1st 1 has been returned (6hrs), 3380864862, which isn't to bad considering all the CPU cores were being used to start with and it's fairly comparable to the time my i5's cores can do them in (the Q6600 cores take about 10hrs each).

Cheers.

After 5 days this experiment is not doing to badly, but once it levels out I'm going to do a bit more fine tuning.

Cheers.

After 5 days this experiment is not doing to badly, but once it levels out I'm going to do a bit more fine tuning.

Cheers.

You also might want to try a later app with the -use_sleep parameter, while also increasing the unroll and ffa_block parameters, that will reduce the CPU usage to minimal levels, while increasing the parameters will make sure it's not a lot slower.

http://setiathome.berkeley.edu/forum_thread.php?id=74075

Claggy

Claggy, once this rig has leveled out as is I'll 1st degrade the driver back to 266.58 just to see what happens with that. After that I'll then start playing around with the parameters. Changing builds will be last thing done. ;-)

Cheers.

OK, things seem to have pretty well leveled out on this experiment so after doing the monthly Windows updates I rolled the driver back to 266.58.

ATM I have 1 card doing an AP while the other is currently doing a MB though once that MB is finished I should hopefully have an AP on both cards and then I'll see what loads I'm getting across the cores.

Once I see what's going on I'll make another change.

Cheers.

After both cards went back to doing AP's the CPU load dropped right off so I changed those original values to 0.5 and now while running AP's on both cards and 3 cores the CPU load is sitting between 76-81%.

So I'll let this go for a few weeks and see how it goes.

Cheers.

Update.

Well after almost 8wks of this test the highest RAC I've seen so far is 10,800 while mainly doing AP's (it's down 1000 atm due to finishing up a lot of MB tasks), but that's nowhere near the 24,000 RAC that this hardware configuration use to get doing MBV6 alone.

Cheers.

... nowhere near the 24,000 RAC that this hardware configuration use to get doing MBV6 alone.

*bites tongue to refrain from commenting on Credit New*

---

Grant
Darwin NT

I should change the title of this thread I suppose.

Cheers.

Well after 4 months of doing mostly AP's on this setup I've just switched it over to MB only now.

Cheers.

Well after 4 months of doing mostly AP's on this setup I've just switched it over to MB only now.

Cheers.

Watch the RAC plummet.
Will be interesting to see where it ends up.

---

Grant
Darwin NT

When Sten-Arne swapped out AP for MB on one of his machines it dropped about 3.5 times, but I think his machine was 100% AP on CPU & GPU. So it will be interesting to see where it lands with this one. I'm going to guess 4500.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

When Sten-Arne swapped out AP for MB on one of his machines it dropped about 3.5 times, but I think his machine was 100% AP on CPU & GPU. So it will be interesting to see where it lands with this one. I'm going to guess 4500.

And he had an AMD GPU, and they tend to be absolutely crap at doing MBv7 in comparison to APv6, so quite a bit of his drop will be because of that, and not NewCredit,

Wiggo on the other hand has two 9800 GT's, they'll be relatively Crap at doing APv6, and a lot better at doing Cuda MBv7, as long as he keeps away from the Cuda22, Cuda42 and Cuda50 builds (drivers permitting),
preferably running the Cuda23 build and not his present Cuda32 build. (the Cuda23 should have a small edge over the Cuda32 build)

Claggy

Actually this test rig is running basically what app's it was when originally in this configuration (except that now it's running AP's on the GPU's as well as MB's until the AP's run out).

I may change the CUDA version later to see what happens, but originally it did better with the 3.2 CUDA version over the 2.3 CUDA version.

This rig may well be shut down again after this trial until my next GPU upgrade and then run the same tests using the GTX 550Ti's that it also had in it (I'm looking at comparisons ATM not what will be best).

Cheers.